Chatbot?

What is a chatbot?

https://medium.com/swlh/what-is-a-chatbot-and-how-to-use-it-for-your-business-976ec2e0a99f

A computer program designed to simulate conversation with human users, especially over the Internet. - Oxford Dictionaries

What is Turing test?

A person talks to both the person and the computer. The goal is to find out who his interlocutor is — a person or a machine.

Chatbot in business

• Find out the weather  
  What’s the weather today?
• Find your way home
  Navigate home.
• Ask Google questions 
  How many kilometres in a mile?
• Play music 
  Play some Jazz music.
• Ask for a joke 
  Tell me a joke.

Chatbot in business

• Answer questions about what is around you
   
• Transferring money
   
• Censorship — Avoids saying certain sensitive words

Chatbot in business

• Healthcare — Symptom Checker Chatbot.
   
• Provide relevant health and triage information based on symptoms;
   
• However, the reliability has been challenged.

Chatbot in business

• Rita — Revolut’s Intelligent Troubleshooting Assistant
   
• For customer support
   
• However, I prefer a real person

1. main components of a chatbot

4 Main components

1. Natural language processing
   - Understanding of the user's message
   - Identifying intents and entities
    
2. Dialog manager
   - Control the flow of the conversation
    
3. Content
   - Scripts of what the chatbot's going to say
    
4. Custom integrations / external data
   - Extra functions: e.g. sending data to DB, prompting external informations to users, sentiment analysis

https://botsociety.io/blog/2018/02/what-is-a-chatbot/

It's a team work to build chatbot

...from scratch

Luckily we have tools available

2. what a chatbot can and cannot do

1. Trying too hard

for passing the Turing test

Trying too hard to pass the Turing test

• Pretend to be human
  - User feels that they are fooled
   
• Handel too many things
  - User feels frustrated
   
• End up passing back the conversation to a human to handle
   
• Better be clear that user is chatting to a bot

2. Lack of clear dialog flow

Lack of Clear dialog flow

• The best commercial bots will try to redirect conversation back to the planed dialog flow
  - aim to collect certain information form user
   
• Handle one thing at a time
   
• Minimize chitchat
  - which is not the chatbot's purpose and strength
   
• User learn how to communicate with the bot

3. Not smart enough

Not Smart Enough

• Not enough training data
  - Bad NLU model
   
• Not enough content diversity
  - Not content generative model
  - Conversation feels dull
   
• Avoid decision-tree types of bots
  - User rather fill in a form

Good Examples

Good Examples

• Good NLP capability
  - multiple entirety extraction
  - adding event in iCalander
  - content extraction in Gmail
   
• Tell users that they are chatting to an assistant (bot, not human)
   
• Be clear what is wanted form the user
   
• Fail gracefully
  - Redirect conversation vs asking for repeat

3. which one to choose

Rasa

• Open Source
   
• Community support
   
• Fully customisation
  - pipeline, model, functionality
   
• Own deployment
   
• Enterprise version available

Watson

• IBM
   
• Slack community
   
• SDK for building with Watson API
   
• Free version: 10,000 free API calls a month
   
• Different pricing for different plans

Dialogflow

• Google
   
• Dialogflow Google Group
   
• SDK and code example available on GitHub
   
• Free Standard Edition
   
• Pay as you go Enterprise Edition

Lex

• Amazon Web Service
   
• Amazon Lex forums
   
• SDKs: Java, JavaScript, Python, CLI, .NET, Ruby on Rails, PHP, Go, and CPP
   
• Limited usage via AWS Free Tier
   
• Or pay only for what you use

Comparison

https://www.globalme.net/blog/lex-vs-dialogflow-vs-watson-vs-rasa

Comparison

My 2 cents

• What is the purpose of your chatbot?
   
• What resources do you have?
   
• How you gonna deploy it?
   
• What skill level do you have?
   
• How much time do you want to invest in it?